repository·master·Indexed 43 days ago
https://github.com/sam31415/timeseriescvA Python package for specialized time series cross-validation designed to prevent data leakage. It provides PurgedWalkForwardCV and CombPurgedKFoldCV classes with a scikit-learn-like API, requiring predictor values, prediction times, and evaluation times to generate train and validation indices.
The timeseriescv package provides two main classes for evaluating machine learning models on time series datasets where samples are tagged with prediction and evaluation times:
PurgedWalkForwardCV: Implements walk-forward cross-validation with purging.CombPurgedKFoldCV: Implements combinatorial cross-validation with purging and embargoing.You can install the timeseriescv package using pip.
pip install timeseriescvThe API is designed to be similar to scikit-learn. The split method is a generator that yields a pair of numpy arrays containing the positional indices for the train and validation sets.
Key differences from scikit-learn:
split method requires X (predictor values), pred_times (prediction times), and eval_times (evaluation times).X, pred_times, and eval_times must be pandas DataFrames or Series that share the same index.